1 /*-
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)uipc_socket2.c 8.1 (Berkeley) 6/10/93
30 */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34
35 #include "opt_param.h"
36
37 #include <sys/param.h>
38 #include <sys/aio.h> /* for aio_swake proto */
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/mbuf.h>
42 #include <sys/mutex.h>
43 #include <sys/proc.h>
44 #include <sys/protosw.h>
45 #include <sys/resourcevar.h>
46 #include <sys/signalvar.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/sx.h>
50 #include <sys/sysctl.h>
51
52 /*
53 * Function pointer set by the AIO routines so that the socket buffer code
54 * can call back into the AIO module if it is loaded.
55 */
56 void (*aio_swake)(struct socket *, struct sockbuf *);
57
58 /*
59 * Primitive routines for operating on socket buffers
60 */
61
62 u_long sb_max = SB_MAX;
63 u_long sb_max_adj =
64 SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */
65
66 static u_long sb_efficiency = 8; /* parameter for sbreserve() */
67
68 static void sbdrop_internal(struct sockbuf *sb, int len);
69 static void sbflush_internal(struct sockbuf *sb);
70
71 /*
72 * Socantsendmore indicates that no more data will be sent on the socket; it
73 * would normally be applied to a socket when the user informs the system
74 * that no more data is to be sent, by the protocol code (in case
75 * PRU_SHUTDOWN). Socantrcvmore indicates that no more data will be
76 * received, and will normally be applied to the socket by a protocol when it
77 * detects that the peer will send no more data. Data queued for reading in
78 * the socket may yet be read.
79 */
80 void
81 socantsendmore_locked(struct socket *so)
82 {
83
84 SOCKBUF_LOCK_ASSERT(&so->so_snd);
85
86 so->so_snd.sb_state |= SBS_CANTSENDMORE;
87 sowwakeup_locked(so);
88 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
89 }
90
91 void
92 socantsendmore(struct socket *so)
93 {
94
95 SOCKBUF_LOCK(&so->so_snd);
96 socantsendmore_locked(so);
97 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
98 }
99
100 void
101 socantrcvmore_locked(struct socket *so)
102 {
103
104 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
105
106 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
107 sorwakeup_locked(so);
108 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
109 }
110
111 void
112 socantrcvmore(struct socket *so)
113 {
114
115 SOCKBUF_LOCK(&so->so_rcv);
116 socantrcvmore_locked(so);
117 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
118 }
119
120 /*
121 * Wait for data to arrive at/drain from a socket buffer.
122 */
123 int
124 sbwait(struct sockbuf *sb)
125 {
126
127 SOCKBUF_LOCK_ASSERT(sb);
128
129 sb->sb_flags |= SB_WAIT;
130 return (msleep(&sb->sb_cc, &sb->sb_mtx,
131 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
132 sb->sb_timeo));
133 }
134
135 int
136 sblock(struct sockbuf *sb, int flags)
137 {
138
139 KASSERT((flags & SBL_VALID) == flags,
140 ("sblock: flags invalid (0x%x)", flags));
141
142 if (flags & SBL_WAIT) {
143 if ((sb->sb_flags & SB_NOINTR) ||
144 (flags & SBL_NOINTR)) {
145 sx_xlock(&sb->sb_sx);
146 return (0);
147 }
148 return (sx_xlock_sig(&sb->sb_sx));
149 } else {
150 if (sx_try_xlock(&sb->sb_sx) == 0)
151 return (EWOULDBLOCK);
152 return (0);
153 }
154 }
155
156 void
157 sbunlock(struct sockbuf *sb)
158 {
159
160 sx_xunlock(&sb->sb_sx);
161 }
162
163 /*
164 * Wakeup processes waiting on a socket buffer. Do asynchronous notification
165 * via SIGIO if the socket has the SS_ASYNC flag set.
166 *
167 * Called with the socket buffer lock held; will release the lock by the end
168 * of the function. This allows the caller to acquire the socket buffer lock
169 * while testing for the need for various sorts of wakeup and hold it through
170 * to the point where it's no longer required. We currently hold the lock
171 * through calls out to other subsystems (with the exception of kqueue), and
172 * then release it to avoid lock order issues. It's not clear that's
173 * correct.
174 */
175 void
176 sowakeup(struct socket *so, struct sockbuf *sb)
177 {
178
179 SOCKBUF_LOCK_ASSERT(sb);
180
181 selwakeuppri(&sb->sb_sel, PSOCK);
182 sb->sb_flags &= ~SB_SEL;
183 if (sb->sb_flags & SB_WAIT) {
184 sb->sb_flags &= ~SB_WAIT;
185 wakeup(&sb->sb_cc);
186 }
187 KNOTE_LOCKED(&sb->sb_sel.si_note, 0);
188 SOCKBUF_UNLOCK(sb);
189 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
190 pgsigio(&so->so_sigio, SIGIO, 0);
191 if (sb->sb_flags & SB_UPCALL)
192 (*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT);
193 if (sb->sb_flags & SB_AIO)
194 aio_swake(so, sb);
195 mtx_assert(SOCKBUF_MTX(sb), MA_NOTOWNED);
196 }
197
198 /*
199 * Socket buffer (struct sockbuf) utility routines.
200 *
201 * Each socket contains two socket buffers: one for sending data and one for
202 * receiving data. Each buffer contains a queue of mbufs, information about
203 * the number of mbufs and amount of data in the queue, and other fields
204 * allowing select() statements and notification on data availability to be
205 * implemented.
206 *
207 * Data stored in a socket buffer is maintained as a list of records. Each
208 * record is a list of mbufs chained together with the m_next field. Records
209 * are chained together with the m_nextpkt field. The upper level routine
210 * soreceive() expects the following conventions to be observed when placing
211 * information in the receive buffer:
212 *
213 * 1. If the protocol requires each message be preceded by the sender's name,
214 * then a record containing that name must be present before any
215 * associated data (mbuf's must be of type MT_SONAME).
216 * 2. If the protocol supports the exchange of ``access rights'' (really just
217 * additional data associated with the message), and there are ``rights''
218 * to be received, then a record containing this data should be present
219 * (mbuf's must be of type MT_RIGHTS).
220 * 3. If a name or rights record exists, then it must be followed by a data
221 * record, perhaps of zero length.
222 *
223 * Before using a new socket structure it is first necessary to reserve
224 * buffer space to the socket, by calling sbreserve(). This should commit
225 * some of the available buffer space in the system buffer pool for the
226 * socket (currently, it does nothing but enforce limits). The space should
227 * be released by calling sbrelease() when the socket is destroyed.
228 */
229 int
230 soreserve(struct socket *so, u_long sndcc, u_long rcvcc)
231 {
232 struct thread *td = curthread;
233
234 SOCKBUF_LOCK(&so->so_snd);
235 SOCKBUF_LOCK(&so->so_rcv);
236 if (sbreserve_locked(&so->so_snd, sndcc, so, td) == 0)
237 goto bad;
238 if (sbreserve_locked(&so->so_rcv, rcvcc, so, td) == 0)
239 goto bad2;
240 if (so->so_rcv.sb_lowat == 0)
241 so->so_rcv.sb_lowat = 1;
242 if (so->so_snd.sb_lowat == 0)
243 so->so_snd.sb_lowat = MCLBYTES;
244 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
245 so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
246 SOCKBUF_UNLOCK(&so->so_rcv);
247 SOCKBUF_UNLOCK(&so->so_snd);
248 return (0);
249 bad2:
250 sbrelease_locked(&so->so_snd, so);
251 bad:
252 SOCKBUF_UNLOCK(&so->so_rcv);
253 SOCKBUF_UNLOCK(&so->so_snd);
254 return (ENOBUFS);
255 }
256
257 static int
258 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS)
259 {
260 int error = 0;
261 u_long tmp_sb_max = sb_max;
262
263 error = sysctl_handle_long(oidp, &tmp_sb_max, arg2, req);
264 if (error || !req->newptr)
265 return (error);
266 if (tmp_sb_max < MSIZE + MCLBYTES)
267 return (EINVAL);
268 sb_max = tmp_sb_max;
269 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES);
270 return (0);
271 }
272
273 /*
274 * Allot mbufs to a sockbuf. Attempt to scale mbmax so that mbcnt doesn't
275 * become limiting if buffering efficiency is near the normal case.
276 */
277 int
278 sbreserve_locked(struct sockbuf *sb, u_long cc, struct socket *so,
279 struct thread *td)
280 {
281 rlim_t sbsize_limit;
282
283 SOCKBUF_LOCK_ASSERT(sb);
284
285 /*
286 * When a thread is passed, we take into account the thread's socket
287 * buffer size limit. The caller will generally pass curthread, but
288 * in the TCP input path, NULL will be passed to indicate that no
289 * appropriate thread resource limits are available. In that case,
290 * we don't apply a process limit.
291 */
292 if (cc > sb_max_adj)
293 return (0);
294 if (td != NULL) {
295 PROC_LOCK(td->td_proc);
296 sbsize_limit = lim_cur(td->td_proc, RLIMIT_SBSIZE);
297 PROC_UNLOCK(td->td_proc);
298 } else
299 sbsize_limit = RLIM_INFINITY;
300 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
301 sbsize_limit))
302 return (0);
303 sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
304 if (sb->sb_lowat > sb->sb_hiwat)
305 sb->sb_lowat = sb->sb_hiwat;
306 return (1);
307 }
308
309 int
310 sbreserve(struct sockbuf *sb, u_long cc, struct socket *so,
311 struct thread *td)
312 {
313 int error;
314
315 SOCKBUF_LOCK(sb);
316 error = sbreserve_locked(sb, cc, so, td);
317 SOCKBUF_UNLOCK(sb);
318 return (error);
319 }
320
321 /*
322 * Free mbufs held by a socket, and reserved mbuf space.
323 */
324 void
325 sbrelease_internal(struct sockbuf *sb, struct socket *so)
326 {
327
328 sbflush_internal(sb);
329 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
330 RLIM_INFINITY);
331 sb->sb_mbmax = 0;
332 }
333
334 void
335 sbrelease_locked(struct sockbuf *sb, struct socket *so)
336 {
337
338 SOCKBUF_LOCK_ASSERT(sb);
339
340 sbrelease_internal(sb, so);
341 }
342
343 void
344 sbrelease(struct sockbuf *sb, struct socket *so)
345 {
346
347 SOCKBUF_LOCK(sb);
348 sbrelease_locked(sb, so);
349 SOCKBUF_UNLOCK(sb);
350 }
351
352 void
353 sbdestroy(struct sockbuf *sb, struct socket *so)
354 {
355
356 sbrelease_internal(sb, so);
357 }
358
359 /*
360 * Routines to add and remove data from an mbuf queue.
361 *
362 * The routines sbappend() or sbappendrecord() are normally called to append
363 * new mbufs to a socket buffer, after checking that adequate space is
364 * available, comparing the function sbspace() with the amount of data to be
365 * added. sbappendrecord() differs from sbappend() in that data supplied is
366 * treated as the beginning of a new record. To place a sender's address,
367 * optional access rights, and data in a socket receive buffer,
368 * sbappendaddr() should be used. To place access rights and data in a
369 * socket receive buffer, sbappendrights() should be used. In either case,
370 * the new data begins a new record. Note that unlike sbappend() and
371 * sbappendrecord(), these routines check for the caller that there will be
372 * enough space to store the data. Each fails if there is not enough space,
373 * or if it cannot find mbufs to store additional information in.
374 *
375 * Reliable protocols may use the socket send buffer to hold data awaiting
376 * acknowledgement. Data is normally copied from a socket send buffer in a
377 * protocol with m_copy for output to a peer, and then removing the data from
378 * the socket buffer with sbdrop() or sbdroprecord() when the data is
379 * acknowledged by the peer.
380 */
381 #ifdef SOCKBUF_DEBUG
382 void
383 sblastrecordchk(struct sockbuf *sb, const char *file, int line)
384 {
385 struct mbuf *m = sb->sb_mb;
386
387 SOCKBUF_LOCK_ASSERT(sb);
388
389 while (m && m->m_nextpkt)
390 m = m->m_nextpkt;
391
392 if (m != sb->sb_lastrecord) {
393 printf("%s: sb_mb %p sb_lastrecord %p last %p\n",
394 __func__, sb->sb_mb, sb->sb_lastrecord, m);
395 printf("packet chain:\n");
396 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt)
397 printf("\t%p\n", m);
398 panic("%s from %s:%u", __func__, file, line);
399 }
400 }
401
402 void
403 sblastmbufchk(struct sockbuf *sb, const char *file, int line)
404 {
405 struct mbuf *m = sb->sb_mb;
406 struct mbuf *n;
407
408 SOCKBUF_LOCK_ASSERT(sb);
409
410 while (m && m->m_nextpkt)
411 m = m->m_nextpkt;
412
413 while (m && m->m_next)
414 m = m->m_next;
415
416 if (m != sb->sb_mbtail) {
417 printf("%s: sb_mb %p sb_mbtail %p last %p\n",
418 __func__, sb->sb_mb, sb->sb_mbtail, m);
419 printf("packet tree:\n");
420 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) {
421 printf("\t");
422 for (n = m; n != NULL; n = n->m_next)
423 printf("%p ", n);
424 printf("\n");
425 }
426 panic("%s from %s:%u", __func__, file, line);
427 }
428 }
429 #endif /* SOCKBUF_DEBUG */
430
431 #define SBLINKRECORD(sb, m0) do { \
432 SOCKBUF_LOCK_ASSERT(sb); \
433 if ((sb)->sb_lastrecord != NULL) \
434 (sb)->sb_lastrecord->m_nextpkt = (m0); \
435 else \
436 (sb)->sb_mb = (m0); \
437 (sb)->sb_lastrecord = (m0); \
438 } while (/*CONSTCOND*/0)
439
440 /*
441 * Append mbuf chain m to the last record in the socket buffer sb. The
442 * additional space associated the mbuf chain is recorded in sb. Empty mbufs
443 * are discarded and mbufs are compacted where possible.
444 */
445 void
446 sbappend_locked(struct sockbuf *sb, struct mbuf *m)
447 {
448 struct mbuf *n;
449
450 SOCKBUF_LOCK_ASSERT(sb);
451
452 if (m == 0)
453 return;
454
455 SBLASTRECORDCHK(sb);
456 n = sb->sb_mb;
457 if (n) {
458 while (n->m_nextpkt)
459 n = n->m_nextpkt;
460 do {
461 if (n->m_flags & M_EOR) {
462 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */
463 return;
464 }
465 } while (n->m_next && (n = n->m_next));
466 } else {
467 /*
468 * XXX Would like to simply use sb_mbtail here, but
469 * XXX I need to verify that I won't miss an EOR that
470 * XXX way.
471 */
472 if ((n = sb->sb_lastrecord) != NULL) {
473 do {
474 if (n->m_flags & M_EOR) {
475 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */
476 return;
477 }
478 } while (n->m_next && (n = n->m_next));
479 } else {
480 /*
481 * If this is the first record in the socket buffer,
482 * it's also the last record.
483 */
484 sb->sb_lastrecord = m;
485 }
486 }
487 sbcompress(sb, m, n);
488 SBLASTRECORDCHK(sb);
489 }
490
491 /*
492 * Append mbuf chain m to the last record in the socket buffer sb. The
493 * additional space associated the mbuf chain is recorded in sb. Empty mbufs
494 * are discarded and mbufs are compacted where possible.
495 */
496 void
497 sbappend(struct sockbuf *sb, struct mbuf *m)
498 {
499
500 SOCKBUF_LOCK(sb);
501 sbappend_locked(sb, m);
502 SOCKBUF_UNLOCK(sb);
503 }
504
505 /*
506 * This version of sbappend() should only be used when the caller absolutely
507 * knows that there will never be more than one record in the socket buffer,
508 * that is, a stream protocol (such as TCP).
509 */
510 void
511 sbappendstream_locked(struct sockbuf *sb, struct mbuf *m)
512 {
513 SOCKBUF_LOCK_ASSERT(sb);
514
515 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0"));
516 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1"));
517
518 SBLASTMBUFCHK(sb);
519
520 sbcompress(sb, m, sb->sb_mbtail);
521
522 sb->sb_lastrecord = sb->sb_mb;
523 SBLASTRECORDCHK(sb);
524 }
525
526 /*
527 * This version of sbappend() should only be used when the caller absolutely
528 * knows that there will never be more than one record in the socket buffer,
529 * that is, a stream protocol (such as TCP).
530 */
531 void
532 sbappendstream(struct sockbuf *sb, struct mbuf *m)
533 {
534
535 SOCKBUF_LOCK(sb);
536 sbappendstream_locked(sb, m);
537 SOCKBUF_UNLOCK(sb);
538 }
539
540 #ifdef SOCKBUF_DEBUG
541 void
542 sbcheck(struct sockbuf *sb)
543 {
544 struct mbuf *m;
545 struct mbuf *n = 0;
546 u_long len = 0, mbcnt = 0;
547
548 SOCKBUF_LOCK_ASSERT(sb);
549
550 for (m = sb->sb_mb; m; m = n) {
551 n = m->m_nextpkt;
552 for (; m; m = m->m_next) {
553 len += m->m_len;
554 mbcnt += MSIZE;
555 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
556 mbcnt += m->m_ext.ext_size;
557 }
558 }
559 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
560 printf("cc %ld != %u || mbcnt %ld != %u\n", len, sb->sb_cc,
561 mbcnt, sb->sb_mbcnt);
562 panic("sbcheck");
563 }
564 }
565 #endif
566
567 /*
568 * As above, except the mbuf chain begins a new record.
569 */
570 void
571 sbappendrecord_locked(struct sockbuf *sb, struct mbuf *m0)
572 {
573 struct mbuf *m;
574
575 SOCKBUF_LOCK_ASSERT(sb);
576
577 if (m0 == 0)
578 return;
579 m = sb->sb_mb;
580 if (m)
581 while (m->m_nextpkt)
582 m = m->m_nextpkt;
583 /*
584 * Put the first mbuf on the queue. Note this permits zero length
585 * records.
586 */
587 sballoc(sb, m0);
588 SBLASTRECORDCHK(sb);
589 SBLINKRECORD(sb, m0);
590 if (m)
591 m->m_nextpkt = m0;
592 else
593 sb->sb_mb = m0;
594 m = m0->m_next;
595 m0->m_next = 0;
596 if (m && (m0->m_flags & M_EOR)) {
597 m0->m_flags &= ~M_EOR;
598 m->m_flags |= M_EOR;
599 }
600 sbcompress(sb, m, m0);
601 }
602
603 /*
604 * As above, except the mbuf chain begins a new record.
605 */
606 void
607 sbappendrecord(struct sockbuf *sb, struct mbuf *m0)
608 {
609
610 SOCKBUF_LOCK(sb);
611 sbappendrecord_locked(sb, m0);
612 SOCKBUF_UNLOCK(sb);
613 }
614
615 /*
616 * Append address and data, and optionally, control (ancillary) data to the
617 * receive queue of a socket. If present, m0 must include a packet header
618 * with total length. Returns 0 if no space in sockbuf or insufficient
619 * mbufs.
620 */
621 int
622 sbappendaddr_locked(struct sockbuf *sb, const struct sockaddr *asa,
623 struct mbuf *m0, struct mbuf *control)
624 {
625 struct mbuf *m, *n, *nlast;
626 int space = asa->sa_len;
627
628 SOCKBUF_LOCK_ASSERT(sb);
629
630 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
631 panic("sbappendaddr_locked");
632 if (m0)
633 space += m0->m_pkthdr.len;
634 space += m_length(control, &n);
635
636 if (space > sbspace(sb))
637 return (0);
638 #if MSIZE <= 256
639 if (asa->sa_len > MLEN)
640 return (0);
641 #endif
642 MGET(m, M_DONTWAIT, MT_SONAME);
643 if (m == 0)
644 return (0);
645 m->m_len = asa->sa_len;
646 bcopy(asa, mtod(m, caddr_t), asa->sa_len);
647 if (n)
648 n->m_next = m0; /* concatenate data to control */
649 else
650 control = m0;
651 m->m_next = control;
652 for (n = m; n->m_next != NULL; n = n->m_next)
653 sballoc(sb, n);
654 sballoc(sb, n);
655 nlast = n;
656 SBLINKRECORD(sb, m);
657
658 sb->sb_mbtail = nlast;
659 SBLASTMBUFCHK(sb);
660
661 SBLASTRECORDCHK(sb);
662 return (1);
663 }
664
665 /*
666 * Append address and data, and optionally, control (ancillary) data to the
667 * receive queue of a socket. If present, m0 must include a packet header
668 * with total length. Returns 0 if no space in sockbuf or insufficient
669 * mbufs.
670 */
671 int
672 sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa,
673 struct mbuf *m0, struct mbuf *control)
674 {
675 int retval;
676
677 SOCKBUF_LOCK(sb);
678 retval = sbappendaddr_locked(sb, asa, m0, control);
679 SOCKBUF_UNLOCK(sb);
680 return (retval);
681 }
682
683 int
684 sbappendcontrol_locked(struct sockbuf *sb, struct mbuf *m0,
685 struct mbuf *control)
686 {
687 struct mbuf *m, *n, *mlast;
688 int space;
689
690 SOCKBUF_LOCK_ASSERT(sb);
691
692 if (control == 0)
693 panic("sbappendcontrol_locked");
694 space = m_length(control, &n) + m_length(m0, NULL);
695
696 if (space > sbspace(sb))
697 return (0);
698 n->m_next = m0; /* concatenate data to control */
699
700 SBLASTRECORDCHK(sb);
701
702 for (m = control; m->m_next; m = m->m_next)
703 sballoc(sb, m);
704 sballoc(sb, m);
705 mlast = m;
706 SBLINKRECORD(sb, control);
707
708 sb->sb_mbtail = mlast;
709 SBLASTMBUFCHK(sb);
710
711 SBLASTRECORDCHK(sb);
712 return (1);
713 }
714
715 int
716 sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control)
717 {
718 int retval;
719
720 SOCKBUF_LOCK(sb);
721 retval = sbappendcontrol_locked(sb, m0, control);
722 SOCKBUF_UNLOCK(sb);
723 return (retval);
724 }
725
726 /*
727 * Append the data in mbuf chain (m) into the socket buffer sb following mbuf
728 * (n). If (n) is NULL, the buffer is presumed empty.
729 *
730 * When the data is compressed, mbufs in the chain may be handled in one of
731 * three ways:
732 *
733 * (1) The mbuf may simply be dropped, if it contributes nothing (no data, no
734 * record boundary, and no change in data type).
735 *
736 * (2) The mbuf may be coalesced -- i.e., data in the mbuf may be copied into
737 * an mbuf already in the socket buffer. This can occur if an
738 * appropriate mbuf exists, there is room, and no merging of data types
739 * will occur.
740 *
741 * (3) The mbuf may be appended to the end of the existing mbuf chain.
742 *
743 * If any of the new mbufs is marked as M_EOR, mark the last mbuf appended as
744 * end-of-record.
745 */
746 void
747 sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n)
748 {
749 int eor = 0;
750 struct mbuf *o;
751
752 SOCKBUF_LOCK_ASSERT(sb);
753
754 while (m) {
755 eor |= m->m_flags & M_EOR;
756 if (m->m_len == 0 &&
757 (eor == 0 ||
758 (((o = m->m_next) || (o = n)) &&
759 o->m_type == m->m_type))) {
760 if (sb->sb_lastrecord == m)
761 sb->sb_lastrecord = m->m_next;
762 m = m_free(m);
763 continue;
764 }
765 if (n && (n->m_flags & M_EOR) == 0 &&
766 M_WRITABLE(n) &&
767 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
768 m->m_len <= M_TRAILINGSPACE(n) &&
769 n->m_type == m->m_type) {
770 bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
771 (unsigned)m->m_len);
772 n->m_len += m->m_len;
773 sb->sb_cc += m->m_len;
774 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
775 /* XXX: Probably don't need.*/
776 sb->sb_ctl += m->m_len;
777 m = m_free(m);
778 continue;
779 }
780 if (n)
781 n->m_next = m;
782 else
783 sb->sb_mb = m;
784 sb->sb_mbtail = m;
785 sballoc(sb, m);
786 n = m;
787 m->m_flags &= ~M_EOR;
788 m = m->m_next;
789 n->m_next = 0;
790 }
791 if (eor) {
792 KASSERT(n != NULL, ("sbcompress: eor && n == NULL"));
793 n->m_flags |= eor;
794 }
795 SBLASTMBUFCHK(sb);
796 }
797
798 /*
799 * Free all mbufs in a sockbuf. Check that all resources are reclaimed.
800 */
801 static void
802 sbflush_internal(struct sockbuf *sb)
803 {
804
805 while (sb->sb_mbcnt) {
806 /*
807 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty:
808 * we would loop forever. Panic instead.
809 */
810 if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len))
811 break;
812 sbdrop_internal(sb, (int)sb->sb_cc);
813 }
814 if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt)
815 panic("sbflush_internal: cc %u || mb %p || mbcnt %u",
816 sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt);
817 }
818
819 void
820 sbflush_locked(struct sockbuf *sb)
821 {
822
823 SOCKBUF_LOCK_ASSERT(sb);
824 sbflush_internal(sb);
825 }
826
827 void
828 sbflush(struct sockbuf *sb)
829 {
830
831 SOCKBUF_LOCK(sb);
832 sbflush_locked(sb);
833 SOCKBUF_UNLOCK(sb);
834 }
835
836 /*
837 * Drop data from (the front of) a sockbuf.
838 */
839 static void
840 sbdrop_internal(struct sockbuf *sb, int len)
841 {
842 struct mbuf *m;
843 struct mbuf *next;
844
845 next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
846 while (len > 0) {
847 if (m == 0) {
848 if (next == 0)
849 panic("sbdrop");
850 m = next;
851 next = m->m_nextpkt;
852 continue;
853 }
854 if (m->m_len > len) {
855 m->m_len -= len;
856 m->m_data += len;
857 sb->sb_cc -= len;
858 if (sb->sb_sndptroff != 0)
859 sb->sb_sndptroff -= len;
860 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
861 sb->sb_ctl -= len;
862 break;
863 }
864 len -= m->m_len;
865 sbfree(sb, m);
866 m = m_free(m);
867 }
868 while (m && m->m_len == 0) {
869 sbfree(sb, m);
870 m = m_free(m);
871 }
872 if (m) {
873 sb->sb_mb = m;
874 m->m_nextpkt = next;
875 } else
876 sb->sb_mb = next;
877 /*
878 * First part is an inline SB_EMPTY_FIXUP(). Second part makes sure
879 * sb_lastrecord is up-to-date if we dropped part of the last record.
880 */
881 m = sb->sb_mb;
882 if (m == NULL) {
883 sb->sb_mbtail = NULL;
884 sb->sb_lastrecord = NULL;
885 } else if (m->m_nextpkt == NULL) {
886 sb->sb_lastrecord = m;
887 }
888 }
889
890 /*
891 * Drop data from (the front of) a sockbuf.
892 */
893 void
894 sbdrop_locked(struct sockbuf *sb, int len)
895 {
896
897 SOCKBUF_LOCK_ASSERT(sb);
898
899 sbdrop_internal(sb, len);
900 }
901
902 void
903 sbdrop(struct sockbuf *sb, int len)
904 {
905
906 SOCKBUF_LOCK(sb);
907 sbdrop_locked(sb, len);
908 SOCKBUF_UNLOCK(sb);
909 }
910
911 /*
912 * Maintain a pointer and offset pair into the socket buffer mbuf chain to
913 * avoid traversal of the entire socket buffer for larger offsets.
914 */
915 struct mbuf *
916 sbsndptr(struct sockbuf *sb, u_int off, u_int len, u_int *moff)
917 {
918 struct mbuf *m, *ret;
919
920 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__));
921 KASSERT(off + len <= sb->sb_cc, ("%s: beyond sb", __func__));
922 KASSERT(sb->sb_sndptroff <= sb->sb_cc, ("%s: sndptroff broken", __func__));
923
924 /*
925 * Is off below stored offset? Happens on retransmits.
926 * Just return, we can't help here.
927 */
928 if (sb->sb_sndptroff > off) {
929 *moff = off;
930 return (sb->sb_mb);
931 }
932
933 /* Return closest mbuf in chain for current offset. */
934 *moff = off - sb->sb_sndptroff;
935 m = ret = sb->sb_sndptr ? sb->sb_sndptr : sb->sb_mb;
936
937 /* Advance by len to be as close as possible for the next transmit. */
938 for (off = off - sb->sb_sndptroff + len - 1;
939 off > 0 && off >= m->m_len;
940 m = m->m_next) {
941 sb->sb_sndptroff += m->m_len;
942 off -= m->m_len;
943 }
944 sb->sb_sndptr = m;
945
946 return (ret);
947 }
948
949 /*
950 * Drop a record off the front of a sockbuf and move the next record to the
951 * front.
952 */
953 void
954 sbdroprecord_locked(struct sockbuf *sb)
955 {
956 struct mbuf *m;
957
958 SOCKBUF_LOCK_ASSERT(sb);
959
960 m = sb->sb_mb;
961 if (m) {
962 sb->sb_mb = m->m_nextpkt;
963 do {
964 sbfree(sb, m);
965 m = m_free(m);
966 } while (m);
967 }
968 SB_EMPTY_FIXUP(sb);
969 }
970
971 /*
972 * Drop a record off the front of a sockbuf and move the next record to the
973 * front.
974 */
975 void
976 sbdroprecord(struct sockbuf *sb)
977 {
978
979 SOCKBUF_LOCK(sb);
980 sbdroprecord_locked(sb);
981 SOCKBUF_UNLOCK(sb);
982 }
983
984 /*
985 * Create a "control" mbuf containing the specified data with the specified
986 * type for presentation on a socket buffer.
987 */
988 struct mbuf *
989 sbcreatecontrol(caddr_t p, int size, int type, int level)
990 {
991 struct cmsghdr *cp;
992 struct mbuf *m;
993
994 if (CMSG_SPACE((u_int)size) > MCLBYTES)
995 return ((struct mbuf *) NULL);
996 if (CMSG_SPACE((u_int)size) > MLEN)
997 m = m_getcl(M_DONTWAIT, MT_CONTROL, 0);
998 else
999 m = m_get(M_DONTWAIT, MT_CONTROL);
1000 if (m == NULL)
1001 return ((struct mbuf *) NULL);
1002 cp = mtod(m, struct cmsghdr *);
1003 m->m_len = 0;
1004 KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m),
1005 ("sbcreatecontrol: short mbuf"));
1006 if (p != NULL)
1007 (void)memcpy(CMSG_DATA(cp), p, size);
1008 m->m_len = CMSG_SPACE(size);
1009 cp->cmsg_len = CMSG_LEN(size);
1010 cp->cmsg_level = level;
1011 cp->cmsg_type = type;
1012 return (m);
1013 }
1014
1015 /*
1016 * This does the same for socket buffers that sotoxsocket does for sockets:
1017 * generate an user-format data structure describing the socket buffer. Note
1018 * that the xsockbuf structure, since it is always embedded in a socket, does
1019 * not include a self pointer nor a length. We make this entry point public
1020 * in case some other mechanism needs it.
1021 */
1022 void
1023 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
1024 {
1025
1026 xsb->sb_cc = sb->sb_cc;
1027 xsb->sb_hiwat = sb->sb_hiwat;
1028 xsb->sb_mbcnt = sb->sb_mbcnt;
1029 xsb->sb_mbmax = sb->sb_mbmax;
1030 xsb->sb_lowat = sb->sb_lowat;
1031 xsb->sb_flags = sb->sb_flags;
1032 xsb->sb_timeo = sb->sb_timeo;
1033 }
1034
1035 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
1036 static int dummy;
1037 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
1038 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_ULONG|CTLFLAG_RW,
1039 &sb_max, 0, sysctl_handle_sb_max, "LU", "Maximum socket buffer size");
1040 SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
1041 &sb_efficiency, 0, "");
Cache object: 9207b797731b95d538db6c0e8df4c085
|